Machine Learning and Artificial Intelligence (Simple Version for Leaders)

I'll start with my apologies for all of you who do this for a living. I'm going to oversimplify and generalize this so non-computer science leaders can understand it a little better.

First, for all practical purposes, Artificial Intelligence does not exist - computers are still dumb (ask Alexa). Computers can’t think like we do (yet). We’re safe from SkyNet for now. And to be fair there is a robot named Sophia that is getting pretty close and she's a passport carrying citizen of Saudi Arabia.

Machine Learning is very real and comes in 2 Flavors:

• Pattern recognition (supervised) - this would be good for recognizing known failure modes (condition-based maintenance), reading text, recognizing corrosion, etc. Patterns can be mapped to known actions (so the computer can advise you what to do next)
• Anomaly detection (unsupervised) - this is good for early warning, recognizing something happening that may result in a problem. This leads to subject matter expert troubleshooting.

Machine Learning gets better and better with more and more data. To get the best we have to normalize data (agree common models) and share data (maybe even with our competitors). This requires a paradigm shift in industry. We will have to find a way to share our data. We will have to align on a common taxonomy or more accurately ontology. What's an ontology you ask? (I did). Taxonomy is a 1:1 hierarchical relationship structure. Ontology is similar concept but allows many:many relationship structure.

Today these tools are very good at friendly problems where the rules are pretty fixed like playing chess or go. They can even be pretty good at running a steady process like air-conditioning, the rules are the laws of physics and if all the equipment is reliable and running the rules stay consistent. They are not good at "wicked" problems. A wicked problem has no fixed rules and the rules sometimes change. Does that sound anything like your problem? Product quality changes, machines break, new technology is introduced with unintended consequences ..... the world is composed of wicked problems (https://en.wikipedia.org/wiki/Wicked_problem). That means that the machine learning can be good at helping you find problems on individual equipment and fairly simple systems, as complexity grows, it becomes more and more difficult for the machine learning to understand the problem. It can only understand problems that have been seen before, not the new ones. Anomoly detection can tell you "something is off" but it takes people to figure out what that is in a wicked system. So don't worry, yet, if your job requires thinking, the machine is not coming for you.